Abbreviation Disambiguation: Experiments with Various Variants of the One Sense per Discourse Hypothesis

Abbreviations are widely used in many languages and disambiguation of abbreviations is critical. In this research, a structured process that attempts to solve the problem of abbreviation ambiguity is presented. Various baseline methods have been explored, including context-related methods and statistical methods. Almost all methods are domain-independent and language independent. The application domain is Jewish Law documents written in Hebrew, which are known to be rich in ambiguous abbreviations. Several implementations of the one sense per discourse hypothesis are used, improving the baseline methods with new variants. Several common machine learning methods have been tested to find a successful integration of the baseline method variants. The best results have been achieved by LIBSVM, with 96.09% accuracy.